摘要
数据挖掘中的聚类分析在给人们带来方便的同时,也凸显了隐私泄露等安全问题,于是隐私保护框架下的聚类分析算法应运而生.考虑到数据集中数据点的每个维度对于数据的重要性或者影响程度均不同,提出了基于距离贡献率的Wk-medoids算法、面向差分隐私的WDPk-medoids算法及面向误差隐私的WEPPk-medoids算法.与原有未加权算法相比,所提算法可降低整个数据点所添加的噪声量、减少加噪数据的失真程度、提高聚类结果的有效性;同时还应用聚类效用评价指标对这三种算法的性能进行了对比分析,为隐私保护框架下聚类挖掘算法如何权衡数据聚类有效性以及隐私保护安全性之间的相互关系提供了参照建议.
Nowadays,clustering analysis in data mining not only brings convenience to people,but also highlights security issues such as privacy disclosure.So the clustering analysis algorithm under the privacy preserving framework came into being.Considering that each dimension of the data point in the dataset has different importance or influence on the data,this paper proposed k-medoids algorithm,WDPk-medoids algorithm oriented to differential privacy and WEPPk-medoids algorithm oriented to error preserving private based on distance contribution rate.Compared with the original unweighted algorithm,the proposed algorithms can reduce the amount of noise added to the whole data point,reduce the distortion of the noisy data,and improve the effectiveness of the clustering results.Besides,the performance of the three algorithms was compared and analyzed by using the clustering utility evaluation index,which provides a reference for the clustering algorithm how to balance the relationship between the data clustering effectiveness and privacy protection securityunder the privacy protection framework.
作者
刘丹青
高瑜
吴振强
LIU Dan-qing;GAO Yu;WU Zhen-qiang(School of Computer,Qinghai Normal University,Xining 810016,China;Computer Science Teaching Department,Dingbian Experimental Middle School,Yulin 718600,China;School of Computer Science,Shaanxi Normal University,Xi’an 710062,China)
出处
《青海师范大学学报（自然科学版）》
2022年第1期4-13,共10页
Journal of Qinghai Normal University(Natural Science Edition)
基金
全国教育科学“十三五”规划2018年度教育部青年课题(EJA180476)
关键词
距离贡献率
差分隐私
误差隐私保护
算法研究
distance contribution rate
differential privacy
error preserving privacy
algorithm research
作者简介
刘丹青(1983-),女,青海西宁人,博士,副教授.研究方向:教育信息化;高瑜(1990-),女,陕西榆林人,硕士,助理讲师.研究方向:网络安全、隐私保护;吴振强(1968-),男,陕西商洛人,博士,教授.研究方向:网络安全、隐私保护.
